<p>Document-level relation extraction (DocRE) involves identifying relations between entities distributed across multiple sentences within a document. Traditional methods often struggle to consider the comprehensive context of the entire document, leading to suboptimal relation classification. To address these challenges, we propose a dual-reasoning enhanced method for DocRE, integrating a two-stage graph convolutional network (GCN) and a Transformer-based module. The graph-based reasoning module GRM leverages a two-stage GCN: the first stage integrates multi-head attention to learn weighted entity representations, while the second stage introduces entity co-occurrence guidance to improve reasoning over the graph. The Transformer-based reasoning module TRM enhances interaction between entity pairs through contextual fusion, enabling fine-grained relation inference. Finally, we integrate the outputs from both modules for relation classification. Extensive experiments on DocRED and Re-DocRED demonstrate that our method achieves strong performance and surpasses most baselines, validating the effectiveness of our dual reasoning design.</p>